A good example of a microprogrammed data processor is described in U.S. Pat. No. 4,173,041, issued on Oct. 30, 1979, to T. J. Dvorak et al. As indicated in this Dvorak et al. patent, the microwords may be located in a separate storage unit called a "control storage" or they may be located in a portion of the processor main storage unit which is set aside for the exclusive use of the microwords. For the case of separate storage units, these units may be of either the read/write type or of the read only type. In smaller data processors, the current trend is to store the microwords in a so-called read-only storage (ROS) unit. In any event, the microword storage units currently used are of the random access type and include the customary addressing circuitry for addressing any desired one of the storage locations in the storage unit.
The current trend is to fabricate the principal circuit sections of a data processor, including the microword control storage section, on large scale integration (LSI) integrated circuit chips. The overall objective is to increase the number of data processing functions that can be provided on a single integrated circuit chip, with the ultimate goal of providing a complete data processor or digital computer on a single chip. Currently, the best technology for packing the largest number of circuits onto a single chip is the so-called metal oxide semiconductor field effect transistor (MOSFET) technology. Such technology makes possible a very high circuit density on an integrated circuit chip. Unfortunately, MOSFET circuits have a somewhat slower operating speed than do circuits fabricated by means of other integrated circuit technologies, such as, for example, the so-called bipolar integrated circuit technology. Thus, it is desirable to find ways to increase the operating speed of data processor mechanisms which employ MOSFET circuits. This is particularly true of the microword control storage section of the data processor because this section provides the elemental control signals for the remainder of the data processor.
Commonly assigned U.S. patent application Ser. No. 350,660, filed Feb. 22, 1982, entitled "Microword Control System Utilizing Multiplexed Programmable Logic Arrays", the inventors being V. S. Moore et al., describes a microword control system wherein the customary random access storage unit and its addressing circuitry are replaced by a plurality of time multiplexed programmable logic arrays (PLA's). Each of the PLA's produces one or more of the microwords needed to execute any given processor instruction. In this system, the access times of at least some of the programmable logic arrays are overlapped so that new microwords become available at a faster rate than for the case of a single programmable logic array.
The multiplexing of the microwords from the different ones of the programmable logic arrays is controlled primarily by the microwords themselves. Each microword is provided with a plural-bit PLA strobe field which is coded to identify the PLA from which the next microword is to be obtained. Two sets of the PLA's may be provided, with each set operating in the manner described. Control point signals derived from the two sets of microwords are time multiplexed to take turns in controlling the operation of the data processor.
The microword control system of the commonly assigned application will now be discussed in detail. Referring now to FIG. 1, there is shown a functional block diagram of a digital computer or digital data processor 10 which is particularly suitable for fabrication on an LSI type integrated circuit chip. This data processor 10 includes a data flow unit 11, a main storage unit 12, an input/output (I/O) unit 13, a control unit 14 and a CPU (central processing unit) clock generator 15. The data flow unit 11 is sometimes called a central processing unit and includes such things as an arithmetic and logic unit (ALU), various hardware registers and counters, a local storage unit and a bus system interconnecting these items with one another. The data flow unit 11 is the unit that does the adding, subtracting, rearranging and other manipulation of the data to produce the desired results.
The control unit 14 controls the operation of the data flow unit 11, the main storage unit 12, and the I/O unit 13 by means of various control point signals which are supplied by way of respective multi-line control buses 16, 17 and 18 to the various functional elements located in these units 11, 12 and 13. Control unit 14 includes a microword generation mechanism for producing a sequence of microwords for each processor instruction to be executed. These microwords, in turn, produce the control point signals which control the elemental operations in the other data processor units.
The data flow unit 11, the I/O unit 13 and the control unit 14 can all be fabricated on one and the same integrated circuit chip.
The user program to be performed by the data processor 10 is initially loaded into the main storage unit 12 from one of the peripheral units (not shown) coupled to the I/O unit 13 by way of an I/O bus 19. This initial loading is accomplished by way of the data flow unit 11. After such initial loading, the user program is performed by reading from the main storage unit 12 in a sequential manner the various processor instructions which make up the user program. Each processor instruction, in its turn, is read out of the main storage unit 12 and loaded into an instruction register 20 located in the data flow unit 11. The processor instruction resident in the instruction register 20, or at least the effective operation code portion thereof, is supplied to the control unit 14 by way of a multi-line bus 21 to identify to the control unit 14 the particular processor instruction to be executed.
In some processors, only a portion of the complete processor instruction may be loaded into the instruction register 20. In such cases, the portion loaded into the instruction register 20 includes at least the "effective" operation code portion of the instruction. By "effective" operation code is meant all the bits in the processor instruction which are needed to uniquely define the kind of operation to be performed by the processor instruction. This does not include operand address bits and length count bits but does include function bits and modifier bits which are needed to complete the definition of the operation to be performed.
Loading of a new processor instruction into the instruction register 20 is accomplished by way of a control point signal which is supplied by way of a control line 22 to a load control terminal of the instruction register 20.
The timing of the operations taking place within the control unit 14 is controlled by the CPU clock generator 15. This clock generator 15 is a two-phase non-overlapping clock generator. It generates the A phase and the B phase clock signals as shown in FIG. 2. Both of these clock signals are of the same frequency and have the same waveshape, the difference being that they are displaced in time relative to one another. The positive pulses of the A clock do not overlap with the positive pulses of the B clock. The amount of non-overlap shown in FIG. 2 is exaggerated for purposes of clarity. These positive pulses represent the active intervals for the two clock signals.
Referring to FIG. 3, there is shown a functional block diagram of microword control system which may be used as the control unit 14 in the FIG. 1 data processor.
The microword control system of FIG. 3 includes first and second sets of PLA mechanisms wherein each PLA mechanism is responsive to the processor instruction to be executed for producing at least one of the microwords needed to execute such instruction. The first set of PLA mechanisms includes a first cycle PLA 201, effective address decode PLA's 202 and 203, and execute decode PLA's 204 and 205. For simplicity, the first set of PLA's will sometimes be referred to by the acronyms P1, PAA, PAC, PXA and PXC, respectively. The second set of PLA mechanisms includes a second cycle PLA 206, effective address decode PLA's 207 and 208 and execute decode PLA's 209 and 210. For simplicity, the second set of PLA's will sometimes be referred to by the acronyms P2, PAB, PAD, PXB and PXD respectively.
Each of these PLA's 201-210 is a dynamic or clocked PLA. For a detailed description of the internal structure thereof, reference may be made to FIG. 8 and the accompanying description in the above-referenced patent application Ser. No. 350,660. Each employs internal C1, C2, C3 and C4 clocking pulses for controlling the internal operations therein.
The microword control unit of FIG. 3 further includes first and second control mechanisms for respectively receiving the microwords from the first and second sets of PLA mechanisms for producing for each microword at least one control point signal for controlling the operation of the data processor. The first control mechanism includes an A-L1 control register 211, an A-L2 control register 212 and an A decoder 213 for receiving the microwords from the first set of PLA's 201-205 via a first PLA output bus 214. The second control mechanism includes a B-L1 control register 215, a B-L2 control register 216 and a B decoder 217 for receiving the microwords from the second set of PLA's 206-210 via a second PLA output bus 218.
The control unit of FIG. 3 also includes first multiplexing circuitry for transferring microwords one at a time from the different ones of the PLA's 201-205 in the first set to the first control mechanism 211-213. This first multiplexing circuitry includes a plurality of selectively operable output buffers (gates) 221-225 for individually coupling the outputs of the different PLA's 201-205 to the input of the A-L1 control register 211 via the PLA output bus 214. These buffers or gates 221-225 are selectively enabled by respective ones of PLA strobes S1, SAA, SAC, SXA and SXC, which PLA strobes are obtained from the output of the A decoder 213. Only one of the strobe signal lines S1, SAA, SAC, SXA and SXC is activated during any given microword cycle, the particular strobe line which is activated being determined by the coding of the PLA strobe field of the microword resident in the A-L2 control register 212.
The control unit of FIG. 3 further includes second multiplexing circuitry for transferring microwords one at a time from different ones of the PLA's 206-210 in the second set to the second control mechanism 215-217. This second multiplexing circuitry includes a plurality of selectively operable output buffers (gates) 226-230 for individually coupling the outputs of the different PLA's 206-210 to the input of the B-L1 control register 215 via the PLA output bus 218. These buffers or gates 226-230 are selectively enabled by respective ones of PLA output strobes S2, SAB, SAD, SXB and SXD which are obtained from the output of the B decoder 217. Only one of these strobe signal lines S2, SAB, SAD, SXB and SXD will be activated during any given microword control cycle, the particular strobe line which is activated being determined by the coding of the PLA strobe field of the microword resident in the B-L2 control register 216 for that particular control cycle.
The microword control system of FIG. 3 further includes third multiplexing circuitry for interleaving the control point signals produced by the first control mechanism 211-213 with the control point signals produced by the second control mechanism 215-217 so that these two control mechanisms take turns in controlling the operation of the data processor. This third multiplexing circuitry includes a multi-stage set of A gates 231 for the A decoder 213 and a multi-stage set of B gates 232 for the B decoder 217. A gates 231 are periodically enabled by the positive A clock pulses from the clock generator 15, while the B gates 232 are periodically enabled by the positive B clock pulses from the clock generator 15. Thus, the control point signals from the A decoder 213 control the data processor during the A clock pulse intervals and the control point signals from the B decoder 217 control the data processor during the B clock pulse intervals. In this manner, the control point signals from the two decoders 213 and 217 are interleaved to take turns in controlling the data processor.
The first and second cycle PLA's 201 and 206 are driven directly from the instruction register 20 in the data processor of FIG. 1. This enables them to react very quickly to new processor instructions. The set of four effective address decode PLA's 202, 203, 207 and 208, on the other hand, receive their processor instruction dependent inputs by way of an effective address encode PLA 233 and a register 234. For simplicity, the encode PLA 233 is assumed to be a static PLA. In response to the processor instruction received via bus 21, encode PLA 233 produces an instruction identification (ID) number having a lesser number of bits than does the processor instruction supplied to the input of the encode PLA 233 by way of the bus 21. This instruction ID number is loaded into the register 234 by an S.0. restart pulse obtained at the appropriate moment from the output of the B decoder 217. The instruction ID number in register 234 is supplied in parallel to a first set of inputs of the AND array potions in each of the effective address decode PLA's 202, 203, 207 and 208.
An effective address sequence counter 235 and a sequence count register 236 supply the sequences of number signals to the second sets of inputs for the AND array portions of the effective address decode PLA's 202, 203, 207 and 208. Sequence counter 235 is reset to a zero condition by the same S.0. restart pulse that loaded the encode PLA output register 234. Depending upon the number of microwords needed from the address decode PLA's, the address sequence counter 235 is incremented one or more times by the SAB strobe pulses obtained from the B decoder 217. The sequence count value in counter 235 is supplied directly to the second sets of inputs of the address decode PLA's 202 and 207. It is supplied by way of register 236 to the second inputs of the other two address decode PLA's 203 and 208. Register 236 is reset to zero by the S.0. restart pulse and is loaded with the sequence count value from counter 235 by the SAC strobe pulse obtained from the A decoder 213.
The set of four execute decode PLA's 204, 205, 209 and 210 are driven in a similar manner by an execute encode PLA 237 and an execute sequence counter 238. The execute encode PLA 237 produces an instruction ID number which is loaded into register 239 by the S.0. restart pulse. This instruction ID number is supplied to the first inputs of each of the execute decode PLA's 204, 205, 209 and 210. The execute sequence counter is reset to a zero value by the S.0. restart pulse and is incremented one or more times by SXB strobe pulses obtained from the B decoder 217. The sequence count numbers from the counter 238 are supplied directly to the second inputs of the execute decode PLA's 204 and 209. They are supplied to the second inputs of the other two execute decode PLA's 205 and 210 by way of register 240. For this embodiment, the execute encode PLA 237 is also assumed to be a static PLA.
The internal C1, C2, C3 and C4 timing pulses for each of the various dynamic PLA's 201-210 are derived from PLA clock signals PC1-PC16 produced by PLA clock logic 242. A different set of four of these PLA clock signals PC1-PC16 is supplied to each of the PLA's 201-210. The first cycle PLA 201, for example, receives the PLA clock signals PC1-PC4. These clock signals PC1-PC4 provide the internal timing signals C1-C4, respectively, needed within the P1 first cycle PLA 201.
The PLA clock logic 242 receives as input signals the various PLA output strobes S1, S2, SAA, SAB, SAC, SAD, SXA, SXB, SXC, and SXD. It further receives as input signals the positive A and B clock pulses from the clock generator 15 and the S.0. restart pulse from the B decoder 217. The PLA clock logic 242 includes therein various combinatorial logic circuits which make use of these various input signals to produce the PLA clock signals PC1-PC16.
Following completion of the effective address calculation sequence of microwords, the execute decode PLA's 204, 205, 209 and 210 become operative to produce a sequence of "execute" microwords PXA, PXB, PXC, PXD, PXA, PXB, . . . , which are, in general, effective to "execute" the processor instruction, that is, to perform the desired operand manipulation or manipulations called for by the processor instruction. After the operand manipulations, the last few "execute" microwords may be used to perform any housekeeping chores that may need to be done to complete the overall microword sequence and to prepare for the coming of the next processor instruction.
The exact number of microwords in the effective address portion of the overall sequence will, in general, be different for different processor instructions. In a similar manner, the exact number of microwords in the execute portion of the overall sequence will also be different for different processor instructions. The last microword in the effective address segment to drive the A decoder 213, in this example, the microword 245, will produce a PLA strobe SXA to select the PXA decode PLA 204 to be the source of the next microword for the A decoder 213. In a similar manner, the last microword in the effective address segment to drive the B decoder 217, in this case, the microword 246, will produce a PLA strobe SXB to select the PXB decode PLA 209 to provide the next microword for the B decoder 217. This accomplishes the switch-over from the effective address segment to the execute segment of the microword sequence. Thereafter, and up until the last two microwords in the execute segment, the microword from each execute decode PLA produces a PLA strobe which selects the next execute decode PLA which is to supply the microword for its respective control register and decoder mechanism. The last A decoder microword in the execute segment, in this example, the microword 247, produces a PLA strobe S1 for selecting the first cycle PLA 201. In a corresponding manner, the last B decoder microword in the execute segment, in this example, the microword 248, produces a PLA strobe S2 for selecting the second cycle PLA 206. In this manner, the last two microwords for the current processor instruction select the sources of the first two microwords for the next processor instruction to be executed.
The distinction between the processor control actions provided by the effective address microwords and the execute microwords is not a hard and fast one. For example, for some processor instructions some of the effective address calculation microwords may, in fact, be performing functions which are more properly classified as execute functions. In other cases, some of the execute microwords may be performing functions which might more properly be grouped with the effective address functions. One reason for not adhering to a strict segregation of functions is to even out the distribution of the microwords so that all of the address and execute decode PLA's may be of approximately the same physical size. This keeps any one of these PLA's from becoming much larger in size than the remainder of the PLA's.
Although the above described microword control system does in fact very significantly increase the operating speed of the data processor in which it is utilized, nevertheless, it would be desirable to still further increase the overall processing speed. Particularly, with the significant increases in processing speed brought about by the above-described microword control system, it has been found that a large portion of the overall operating time of the data processor employing the microword control system is taken up by internally generated, that is, system-initiated, control operations. Examples of such operations are stop, wait and load, error check, program check, machine check, stop-on-error, check-restart, interrupt-request, reset-interrupt, I/O interrupt, power failure, or over-temperature shut-down, system reset, initial program loading (IPL) and power reset. The processing of such control operations has heretofore been handled in a conventional manner. For instance, a table of instructions may be stored in a read-only memory for each internally generated control operation, the set of instructions accessed each time that one of the internally generated control operations is to be executed, and the instructions decoded one-by-one to generate the control point signals necessary for carrying out the desired control operation. As mentioned above, this manner of processing internally generated control operations is quite time consuming, significantly lowering the overall average processing speed of the data processor.